1

DEVOICING OF WORD-INITIAL STOPS: A CONSEQUENCE OF THE

FOLLOWING VOWEL?

Daniel Pape 1, Christine Mooshammer 2, Phil Hoole 3 & Susanne Fuchs 1

1 ZAS Berlin, Germany

2 Christian-Albrechts-Universität zu Kiel, Germany

3 LMU München, Germany

ABSTRACT: The aim of the current study is to investigate the contextual conditions of

devoicing of phonologically voiced stops. Therefore articulatory and acoustical data of four

male speakers were recorded by means of EMMA and EPG. Devoicing was observed more

frequently for the velar stops than for the bilabials. The highest occurrence of devoicing was

observed when the voiced stop was followed by a low or mid vowel. To test whether

articulatory positions are affected by the identity of the following vowel ANOVAs were

computed. All subjects showed significant effects on positional data varying with place of

articulation of the stop. Percentage of devoicing was significantly correlated with vertical and

horizontal tongue positions for the velar and with the vertical jaw position for both stops.

Stepwise regression models were computed to achieve an objective measure for the relevance

of the measured parameters. We assume that in German movement economy, i.e.

coarticulation, is more important than the maintenance of voicing during the closure, which is

in agreement with the view that the voicing distinction in German is primarily produced by a

longer VOT for the voiceless stops.

2

1. INTRODUCTION

Devoicing of stops can be attributed to the fact that due to the accumulation of air behind the

closure the transglottal pressure drop decreases and the vocal folds stop vibrating (see e.g.

Ohala and Riordan, 1980).

Accordingly devoicing of stops is generally a natural consequence of an oral closure,

assuming the velar port is closed and the speaker does not produce an active and/or passive

mechanism to overcome the transglottal pressure drop. A passive mechanism for this

maintenance of voicing would be soft tissue compliance, e.g. for the cheeks during the

production of bilabials. A possible active strategy to counteract the effect of air accumulation

behind the closure is to enlarge the intraoral cavity. This so called cavity enlargement was

investigated by Westbury (1983) who found that depending on the place of articulation

tongue, jaw, larynx and the soft palate can contribute to an increase of oral volume.

Thus there seems to be a trade-off between consequences of the physical properties of our

speech apparatus, i.e. devoicing as a consequence of oral closure on the one hand and

language dependent demands related to the voicing contrast on the other hand: In Romance

languages voicing is often found to be maintained throughout the complete stop closure,

therefore mechanisms like cavity enlargement should be necessary. In contrast, for Germanic

languages the distinction between phonologically voiced and voiceless stops is mainly based

on differences in aspiration, i.e. no additional strategy for the maintenance of voicing would

be required. This language contrast was experimentally examined by means of “voicing

profiles” by Shih and Möbius (1999), which trace frame-by-frame voicing status from the

3

beginning of a stop closure until the stop release. They found a language dependent contrast

for Italian and Spanish versus German with almost no devoicing for the former languages and

lower devoicing probability for the latter. Fischer-Jørgensen (1968) found similar results

examining the voicing patterns of French versus Danish stops.

However, in both language families velar stops are more prone to devoicing in comparison to

bilabial stops. This phenomenon can be attributed to aerodynamics: For velar stops the back

cavity is rather small and probably allows only restricted use of enlargement strategies

whereas in bilabial stops such mechanisms could be applied rather easily. Maddieson (2003)

provided evidence that the “missing /g/” patterns in the phoneme inventories of the sounds of

the world’s languages occur rather frequently; this he attributed to the morphology of the

vocal tract in combination with aerodynamic factors.

Several factors have been found to influence devoicing of phonologically voiced stops:

1) Place of articulation: Ohala and Riordan (1980) found empirically that velar stops are more

often subject to devoicing due to less volume behind the point of constriction. This limits

their capacity for passive enlargement which is necessary to keep the pharyngeal pressure

low. Keating et al. (1983) found that the duration of voicing into closure varies with place of

articulation in English and Swedish.

2) Position in utterance: To examine the likelihood of voiced and devoiced stops in different

positions in an utterance, Westbury and Keating (1985) computed the different aerodynamic

conditions given by different positions of a stop in an utterance. They found that from an

aerodynamic point of view a voiced stop is more likely to be produced in medial position.

4

However, in utterance initial and final position aerodynamic demands are more likely to

produce a voiceless stop.

3) Voicing status of context: Shih and Möbius (1999) found strong contextual influences on

the devoicing patterns of stops in different languages, i.e. the devoicing of phonologically

voiced German stops was dependent on whether the preceding context was voiced (vowels

and sonorants) or voiceless (voiceless stops and voiceless fricatives), with lower percentage

of devoicing if the preceding phone was voiced).

4) Vowel context: Ohala and Riordan (1980) observed that stops coarticulated with high

vowels permitted voicing to continue longer than those coarticulated with low vowels, due to

the enlarged pharyngeal cavity for high vowels.

5) Stress: Keating et al. (1983) found that stress increased the duration of closure voicing for

Swedish.

6) Duration of the stop: The longer the stop closure the higher the probability that voicing will

cease. Ohala and Riordan (2003) found in a vented valve experiment controlling the oral air

pressure artificially that voicing during the stop production could not be maintained for longer

than about 60ms. Kawahara (2004) examined Japanese singleton stops in comparison to

geminates and found that voiced singleton stops showed voicing for almost 100 percent of

closure duration whereas for voiced geminates on the average only around 30 to 40 percent of

the stop closure maintained voiced.

5

The general aim of the current investigation is to study the dependency of devoicing effects in

the following vowel. Therefore, we extend the work of Ohala and Riordan (1980) to a greater

variety of vowel contexts, i.e. to the whole German vowel inventory. The second aim is to test

their hypothesis that coarticulatory influences cause the vowel-specific distribution of

devoicing occurrence by means of articulatory and acoustic measurements. Therefore, we

conducted a combined EMMA, EPG and acoustic experiment to investigate the causes for

devoicing of phonologically voiced word-initial stops in German.

In particular we were interested in (1) whether the patterns of devoicing in German resembled

the patterns in American English, (2) whether the articulatory configuration at the onset of the

consonantal constriction is already influenced by the following vowel and (3) whether the

patterns of devoicing can be explained by these anticipatory effects.

2. METHOD

2.1. Experimental setup

We investigated tongue and jaw movements together with tongue-palate contact patterns by

means of synchronized EPG (Reading EPG3), EMMA (AG100, Carstens Medizinelektronik)

and acoustic recordings of four male subjects (CG, DF, JD, RW). Four sensors were attached

mid-sagittally to the tongue spaced equally from 1 to 5 cm behind the tongue tip, one to the

jaw (lower incisors) and one to the lower lip. Two sensors, one at the bridge of the nose and

the other at the upper incisors, served as reference coils to compensate for helmet movements

6

during the recording session. The audio signal was simultaneously recorded on DAT. The

final sampling frequency for the articulatory data was 200Hz and 16kHz for the acoustic data.

The speech material consisted of nonsense words /Vk/ for the velars and /bVp/ for the

bilabials, where V consisted of the 14 German tense and lax vowels

/i y u e ø o œ a/. The target words were embedded in the carrier phrase “Sage

__ bitte.” (“Say __ please.”). Since the devoicing of stops occurs only rarely in word-medial

position we chose the word initial position to examine devoicing. Obviously a word boundary

preceding the stop occurred and could introduce problems with the length of the stop

duration.

Each sentence was repeated 10 times except for speaker RW whose sensors came off after 8

repetitions.

2.2. Acoustic and articulatory measurements

Following our intention to study the strength of the vowel specific influence on articulatory

positions during the stop, a reference point was chosen early during the closure of the stop.

The reasons for this choice were (1) that we assumed that if the patterns of devoicing can be

explained by anticipation then articulatory positions should differ before devoicing occurs

and (2) it is guaranteed that for all tokens voicing is still maintained at this time-point.

Therefore we chose the acoustically defined onset of the stop closure. This onset is usually

labelled as the offset of higher formants (Klatt, 1975), preferably the second formant.

Two problems occurred during labelling the data: First, it was found that frequently the offset

of the different higher formants did not occur at the same acoustical landmark, therefore

7

introducing strong measurement variability. Secondly, even when concentrating on the offset

of the second formant, one speaker showed strong nasalization during the preceding vowel,

the speakers anticipatory strategy to maintain voicing throughout the following closure (this

nasalization was clearly audible and confirmed by informal perceptual tests). This frequently

resulted in a strong weakening of the second formant of the vowel to be measured, due to the

interference with resonances of the nasal cavity (Stevens, 1998). The resulting increased

variability led us to a different labelling technique in order to get a more reliable

measurement: A relevant decrease of the sound intensity was operationally labelled as –6dB,

measured from the point of maximal intensity of the preceding vowel. Since there was no

aerodynamic (or linguistic reason) to assume that at this acoustic landmark the speaker is

actively decreasing the acoustic intensity of the glottal output, the –6dB decrease of global

intensity indicates the beginning of the stop closure and was in about 95% of the cases

identical with the offset of the second formant of the vowel. Figure 1 indicates the procedure.

The intensity decrease was automatically measured using a script with the standard intensity

settings of the software PRAAT (Boersma and Weenink 1999).

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Figure 1 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

At this acoustically defined landmark of closure onset we measured the horizontal and

vertical positions of the tongue dorsum (TDORS), tongue back (TBACK), jaw and lower lip

(LLIP) sensors. From the EPG data the centre of gravity index (COG) and the percentage of

8

contacts in the posterior palatal region (POST) were calculated by using the formulae given

in Gibbon and Nicolaidis (1999).

The stop was labelled as devoiced if there was no visible periodicity in the oscillogram for

more than one glottal period (about 10ms for male subjects) preceding the burst.

3. RESULTS

3.1. Occurrence of devoicing

Figure 2 shows the percentages of devoicing for all speakers split by place of articulation and

following vowel. As was expected the velar stop is more often subject to devoicing than the

bilabial stop (46.3 per cent vs. 26.4 per cent). The percentage of devoicing clearly increases

with decreasing vowel height, e.g. the bilabial stop was more often devoiced when followed

by the mid and low vowels /, a, , / compared to the high vowel /i, y, u, e/. These

findings are generally in agreement with previous studies.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Figure 2 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Table 1 shows the percentage of stop devoicing for each vowel, split by speaker and stop.

Data of each speaker show similar relationships between devoicing and vowel height, but the

overall amount of devoicing varies speaker-dependently, e.g. speaker DF is more prone to

9

devoice the bilabial than speaker CG. Speaker JD’s devoicing pattern for the bilabial is

exceptional, with almost no instances of devoicing. This speaker avoids devoicing by

prenasalizing the bilabial but not the velar stop.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Table 1 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

3.2. Anticipatory effects on articulatory positions at consonantal closure onset

The vowel-specific distribution of devoicing suggests that the following vowel influences the

volume of the oral cavity during occlusion. To test whether tongue, jaw and lip positions are

also affected by the identity of the following vowel at the moment of the acoustically defined

consonant onset ANOVAs were calculated with vowel identity as independent variable and

positions of the articulators and EPG-measures as dependent variables, split by speaker and

consonant. Since vowel identity is a factor with 14 levels we used the Scheffé post-hoc test

which is known to be highly conservative which means that very large differences between

means are needed to detect different groups. Therefore in some cases the ANOVA gives

significant effects where the Scheffé test shows no significantly different groups. These cases

are marked in Table 2 by brackets around the asterisks, whereas when both the ANOVA and

the Scheffé test yielded significant effects and groups the F-values then they are printed in

bold. In the following section anticipatory effects will only be considered for articulators if

the Scheffé test showed significant groups too.

10

For the EPG measures the COG was insensitive for vowel context (except for speaker JD) but

the POST increased significantly for all speakers and both consonants comparing front high

vowels with low back vowels. At the onset of the velar the position of the most posterior

tongue sensor (TBACK) was not affected by the vowel but the vertical positions of the more

fronted sensor (TDORS) varied significantly with vowel height for three speakers (not

significant for RW). Jaw positions for velars tended to be higher for following closed vowels

for speaker CG and higher for rounded vowels for speaker DF (both almost significant).

In contrast to the velar, for the bilabial both tongue sensors varied significantly according to

the following vowel in the expected direction (the TDORS sensor was affected more

consistently). The jaw tended to be significantly more closed when the following vowel was

rounded for speakers DF and JD (for speaker CG the jaw was influenced by vowel hight but

not by rounding). The horizontal lip position at the onset of the bilabial was affected by

rounding for three speakers but significant only for speaker JD, the speaker who maintained

voicing most effectively for the bilabials. This speaker also showed a peculiar pattern for the

vertical lip position, with lower values for following rounded vowels, whereas the other

speakers tended to lower the lower lip for unrounded vowels (not significant).

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Table 2 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

To summarize the results so far, significant vowel effects could be detected at the onset of the

consonant. As was expected those articulators which do not directly contribute to the

11

production of the stop vary most consistently with vowel identity, e.g. lip positions for the

velars and tongue positions for the bilabial. For the jaw whose task for producing a bilabial

and velar closure is rather unclear we assume a helping function which is probably more

pronounced in the bilabial stop. For both contexts the vertical jaw position varied for all

speakers (except RW) but no consistent pattern could be extracted: For speaker CG the jaw

height is influenced by vowel height whereas for speaker DF the jaw is elevated for rounded

vowels (only for the bilabials for speaker JD). The presumably relevant articulator, e.g. the

lower lip for bilabials and both tongue sensors for the velars, also varied for three speakers

but the influence was weaker and less consistent compared to the irrelevant articulators.

3.3. Relationship Devoicing – Articulatory Positions

To analyse which of the articulators might have an influence on the occurrence of devoicing

we computed correlations between positional data and the percentage of devoicing calculated

over 10 repetitions of each item. Since it is well known that the occurrence of devoicing is

also influenced by closure duration (see Introduction) we added this variable (DurC) to our

analyses. Table 3 shows the correlation coefficients and the level of significance.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Table 3 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

12

For the bilabial stop the devoicing pattern can only be related to the vertical jaw position for

two speakers with a highly significant negative correlation, i.e. the lower the jaw position the

higher the percentage of devoicing. This is also captured by Figure 3 (in the left panels),

which shows scatterplots of the percentage of devoicing and the averaged vertical jaw

position, split by speaker and consonant. The jaw positions of speakers CG and DF also

varied significantly with vowel height. Speakers JD and RW showed no significant

correlations at all for the bilabial. For speaker JD obviously this can be attributed to the very

rare instances of devoicing. Lip rounding (LLIPX) was never significantly related to the

percentage of devoicing. The duration of the bilabial (DurC) was positively correlated for

two speakers, i.e. the longer the stop the higher the likelihood for devoicing.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Figure 3 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

For the velar stop the devoicing pattern is also related to the vertical jaw position for three

speakers, for CG with a highly significant correlation. For two speakers the POST and for

speaker RW the COG were significantly negatively correlated with the percentage of

devoicing. The vertical tongue sensors showed a negative correlation, i.e. the lower the

tongue, the higher the percentage of devoicing (not significant for speaker CG). For the two

posterior tongue sensors, which are assumed to capture the velar articulator best, oppositional

patterns could be observed for speakers CG and JD on the one hand and speaker DF on the

13

other hand. The latter fronted the place of articulation before front vowels. Because of the

curved shape of the palate the tongue was also higher before front vowels. Furthermore

speaker DF tended to devoice /g/ more frequently when the stop was followed by a back

vowel which was not the case for speakers CG and JD. The latter two speakers showed no

fronting of the velar with a following front vowel but significant negative correlations

between horizontal tongue positions and devoicing, i.e. the more fronted the tongue the more

frequently devoicing occurred. No significant correlations between devoicing and horizontal

tongue positions could be found for speaker RW.

To achieve an objective measure of the relevance of the measured parameters for the

occurrence of devoicing SPSS stepwise regression models were computed with percentage of

devoicing as dependent variable and articulatory positions of tongue back, tongue dorsum,

jaw and lower lip sensors, the EPG measures POST and COG as well as the stop duration as

independent variables. Table 4 shows the extracted regression models with the predictors

selected by stepwise regression models, explained variance (R2), the F values and the

probability. As can be seen no model could be extracted for speaker JD for the devoicing

pattern of bilabials. Closure duration (DurC) only played a significant role for the bilabials

(speakers DF and RW) but not for the velars. The inclusion of the vertical jaw component

improved the prediction considerably for speakers DF and RW whereas for speaker CG the

vertical jaw position was actually the only variable that met the criterion for inclusion

(F>3.84) and explains about 45% of the variance. The occurrence of devoicing of velar stops

was best predicted by the vertical jaw position for speakers CG and RW. In two cases the

horizontal jaw position was included in the models. The inclusion for speaker CG can be

attributed to a suppression effect of JAWX (Bortz 1979) because it did not correlate with the

criterion variable but significantly improved the model due to the high correlation between

14

JAWX and JAWY (r= -.586*). This was not the case for speaker JD whose pattern of

devoicing significantly varied with jaw retraction (see Table 3). The main predictor variable

for this speaker was the horizontal tongue dorsum position. Speaker DF, as can be seen in

Figure 3, produced the velar stop with a higher jaw position when followed by a rounded

vowel. Therefore the jaw was not selected as a predictor variable for devoicing but the

vertical tongue dorsum position.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Insert Table 4 here

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

4. CONCLUSION AND DISCUSSION

In this study, we investigated the occurrence of devoicing in phonologically voiced stops and

its dependency on the following vowel in German. We measured articulatory data (EMMA

and EPG) at the acoustically defined closure onset. Dependent on the place of articulation of

the stop, devoicing was more frequently found for the velar with 46.3 per cent compared to

the bilabial with 26.4 per cent. In accordance with earlier studies on American English

(Ohala and Riordan, 1980), we found that the percentage of stop devoicing increases with

increasing vowel openness. Ohala and Riordan attributed this devoicing pattern to the

decreased pharyngeal volume in anticipation of following lower vowels. The presence of

anticipatory effects at the onset of the stop closure was confirmed by ANOVAs and varied

with place of articulation of the stop.

15

Correlation between articulatory data and percentage of devoicing revealed significant

correlations for the vertical jaw position for both stops. In the stepwise regression model this

articulator was selected as predictor variable for all speakers who showed a devoicing pattern

for the bilabial and two speakers for the velar. For the velar significant correlations with

vertical and horizontal tongue positions were found for three speakers. Closure duration

showed only a significant effect on the percentage of devoicing for the bilabial for two of the

four speakers. Since passive compliance of the walls plays a greater role for bilabials (see

Ohala, 1983) occurrences of devoicing might be more strongly affected by closure duration,

i.e. the cavity enlargement due to lax cheeks provides a sufficient pressure drop only for

shorter closure durations. This possibility of cavity enlargement does not exist for velars.

Obviously not all relevant factors explaining devoicing patterns were captured by our

experimental set-up, which can be seen in the low explained variances in the stepwise

regression models. Aerodynamic factors such as the volume of pharyngeal cavity, transglottal

and intraoral pressure were not captured nor were positions of articulatory structures such as

larynx height or velum which contribute to the size of the oral cavity (see Westbury 1983).

Ohala and Riordan’s hypothesis that pharyngeal volume plays the major role for vowel-

specific devoicing patterns can only be tested indirectly with our data: As was found by

Tiede (1996) tongue dorsum height and pharyngeal volume are highly positively correlated

for English vowels. In our regression models the tongue dorsum height was relevant only for

one speaker in velar context whereas the jaw was one of the major factors. One possible

explanation is that the jaw captures vowel height more consistently compared to the two

tongue sensors which are strongly influenced by vowel frontness (e.g. /i:/ is produced with a

higher tongue dorsum position than /u:/ due to the shape of the palate).

16

Another vowel-specific anticipatory effect which also enlarges the cavity and therefore might

be involved in maintenance of voicing is larynx height, which has been shown to be

significantly lower for rounded vowels (see for German e.g. Hoole and Kroos 1998). The

hypothesis here would be that if the larynx height is adjusted already at the beginning of the

stop then stops followed by rounded vowels should be less frequently devoiced compared to

unrounded vowels with the same vowel height. This pattern of devoicing was only found for

one speaker (DF) in bilabial context (see Table 1) whose jaw position was also mainly

dependent on rounding (see Figure 3). Since this pattern was only found for one speaker and

one context larynx height does not seem to play a major role. The other possibility to control

vocal tract length is lip rounding but this strategy was not used by our speakers (e.g. no

significant correlations between devoicing and lip positions for the bilabials, as can be seen

in Table 3).

However, even with our limited data set we are tempted to conclude that cavity enlargement

does not seem to play a major role in the production of German stops. Our results are in

accordance with the results of Jessen (2001) and others who stated that in Germanic

languages other features for voiced/voiceless stop distinction, mainly aspiration duration, are

of greater importance than the maintenance of voicing throughout the complete stop closure.

Even though anticipatory effects of the following vowel on the occurrence of stop devoicing

vary speaker-dependently, requirements for economy of movement play a more important

role than the maintenance of voicing in German.

5. ACKNOWLEDGEMENTS

17

This work was supported by a grant from the German Research Council GWZ 4/8-1, P.1. We

would like to thank Jörg Dreyer at the ZAS for technical setup and maintenance, especially

for the realization of the synchronized recording of EMMA, EPG and acoustic data. Thanks

to Christian Geng for advises regarding the regression statistics.

6. REFERENCES

Boersma, P. and Weenink, D. (1999). “PRAAT, a system for doing phonetics by computer”

Report of the Institute of Phonetic Sciences of the University of Amsterdam, 132-182.

Bortz, J. (1979). Lehrbuch der Statistik – Für Sozialwissenschaftler. Berlin: Springer Verlag

Gibbon, F. and Nicolaidis, K. (1999). Palatography. In W.J. Hardcastle and N. Newlett

(Eds.), Coarticulation (pp 229-245), Cambridge: Cambridge University Press.

Fischer-Jørgensen, E. (1968). Voicing, Tenseness and Aspiration in Stop Consonants, with

Special Reference to French and Danish. ARIPUC, 3, 63-115.

Hoole, P. and Kroos, C. (1998). Control of larynx height in vowel production. Proceedings of

the 5th International Conference on Spoken Language Processing, 2, 531-534.

Jessen, M. (2001). Phonetic implementation of the distinctive auditory feature [voice] and

[tense] in stop consonants. In Hall, T.A. (Ed.), Distinctive feature theory (pp 237-294).

Berlin: Mouton De Gruyter.

Kawahara, S. (2004). An Acoustic and perceptual study of Japanese voiced geminates.

Manuscript for UMASS Mini-Conference.

18

Klatt, D.H. (1975). Voice onset time, frication, and aspiration in word-initial consonants

clusters. Journal of Speech and Hearing Research, 18, 686-706.

Keating, P., Linker, W. and Huffman, M. (1983). Patterns in allophone distribution for voiced

and voiceless stops. Journal of Phonetics, 11, 277-290.

Maddieson, I. (2003). Phonological typology in geographical perspective. Proceedings of the

15th ICPhS in Barcelona, 719-722.

Shih, C. and Möbius, B. (1999). Contextual effects on consonantal voicing profiles: A cross-

linguistic study. Proceedings of the 14th ICPhS in San Francisco, 2, 989-992.

Stevens, K.N. (1998). Acoustic phonetics (p. 512). Cambridge: The MIT Press.

Tiede, M.K. (1996). An MRI-based study of pharyngeal volume contrasts in Akan and

English. Journal of Phonetics 24 399-421

Ohala. J. and Riordan, C. (1980). Passive vocal tract enlargement during voiced stops. Report

of the Phonological Laboratory UC Berkeley 5, 78-87.

Ohala, J. (2003). Effects on speech of introducing aerodynamic perturbations. Proceedings of

the 15th ICPhS in Barcelona, 2913-2916.

Westbury, J.R. (1983). Enlargement of the supraglottal cavity and its relation to stop

consonant voicing. Journal of the Acoustical Society of America, 73, 1322-1336.

Westbury, J.R. and Keating, P. (1985). On the naturalness of stop consonant voicing. Journal

of Linguistics, 22, 145-166.

19

FIGURES AND TABLES

Figure 1: Procedure of the acoustic labelling: The upper plane shows the oscillogram, the

middle plane the spectrogram and the lower plane shows the intensity contour. The left black

bar (t1) indicates the landmark of the maximal intensity of the vowel. The right bar (2)

indicates the 6dB intensity decrease measured from the point of maximal intensity.

Figure 2: Percentage of devoicing, shown for each vowel and the contexts /b/ and /g/. The

vowels are ordered by their phonological vowel height.

20

Figure 3: Scatterplots of the percentage of devoicing and the vertical jaw position at

consonant onset of bilabials (left) and velars (right) for means of each speaker.

Table 1: Percentage of stop devoicing split by speaker and place of articulation depending on

the following vowel.

Bilabial stop Velar stop Vowel CG DF JD RW CG DF JD RW i 0.00 20.00 0.00 0.00 0.00 50.00 0.00 25.00 y 0.00 18.18 0.00 12.50 20.00 66.67 10.00 37.50 u 10.00 0.00 10.00 50.00 20.00 63.64 20.00 37.50 20.00 60.00 0.00 37.50 11.11 63.64 40.00 44.44 10.00 70.00 0.00 12.50 10.00 80.00 25.00 37.50 20.00 18.18 0.00 12.50 0.00 63.64 10.00 55.56 e 0.00 27.27 20.00 12.50 0.00 60.00 22.22 37.50 ø 0.00 18.18 0.00 0.00 10.00 80.00 50.00 12.50 o 0.00 18.18 0.00 37.50 40.00 100.0 10.00 50.00 50.00 100.0 10.00 62.50 20.00 81.82 66.67 77.78 œ 20.00 70.00 0.00 62.50 40.00 90.91 30.00 50.00 30.00 72.73 0.00 62.50 70.00 100.0 80.00 37.50 40.00 70.00 10.00 62.50 80.00 81.82 40.00 62.50 a 70.00 100.0 0.00 100.0 60.00 100.0 33.33 50.00

21

Table 2: ANOVAS showing the F-values, the degrees of freedom and the significance levels

(*** p<0.001; ** p<0.01; * p<0.05) for each speaker split by bilabial and velar context. The

asterisks in brackets indicate cases where the ANOVA yielded significant effects, but the

more conservative Scheffé-post-hoc test could not separate significant different groups. F-

values printed in bold indicate significant different vowel groups obtained by Scheffé post-

hoc test.

Bilabial stop CG DF JD RW F(13,126): F(13,123): F(13,126): F(13,98): POST 6.8 *** 17.6 *** 5.0 *** 34.6 *** COG 0.4 0.6 3.7 (***) 2.3 (**) TBACKX 5.1 (***) 8.5 *** 12.9 *** 4.4 *** TBACKY 5.2 *** 7.7 *** 2.0 (*) 3.1 (***) TDORSX 15.3 *** 7.5 *** 5.6 *** 10.0 *** TDORSY 11.0 *** 23.5 *** 7.3 *** 8.8 *** JAWX 0.3 0.6 1.8 1.5 JAWY 4.1 *** 5.2 *** 10.0 *** 1.4 LLIPX 3.3 (***) 1.5 11.4 *** 1.6 LLIPY 2.5 (***) 6.1 *** 9.8 *** 1.0 Velar stop CG DF JD RW F(13,119): F(13,133): F(13,120): F(13,101): POST 3.6 *** 7.5 *** 19.0 *** 5.4 *** COG 0.8 0.6 12.7 ** 1.7 TBACKX 1.7 0.8 2.0 (*) 0.8 TBACKY 2.1 (*) 6.1 *** 1.0 1.1 TDORSX 4.3 *** 1.9 (*) 1.0 2.3 (***) TDORSY 5.1 *** 18.1 *** 11.0 *** 3.2 (***) JAWX 2.0 (*) 3.6 (***) 1.7 1.8 JAWY 4.2 (***) 5.2 (***) 3.0 (***) 3.0 (***) LLIPX 47.2 *** 27.2 *** 37.1 *** 13.7 *** LLIPY 13.3 *** 31.9 *** 4.3 *** 10.4 ***

22

Table 3: Correlation coefficients between devoicing and articulatory measurement points.

Significant values (p=.05) are marked with a grey cell background. Highly significant values

(p=.01) are bold printed.

Bilabial stop Velar stop CG DF JD RW CG DF JD RW TBACKX 0.291 -0.113 -0.273 0.173 -0.577 0.593 -0.483 -0.386 TDORSX 0.354 -0.050 -0.345 0.353 -0.167 0.544 -0.664 0.090 JAWX 0.214 0.276 -0.096 -0.258 0.009 -0.250 0.610 -0.017 LLIPX 0.433 0.156 0.175 -0.133 0.053 -0.234 0.348 0.366 TBACKY -0.399 -0.129 -0.168 -0.499 -0.425 -0.638 -0.571 -0.292 TDORSY -0.478 -0.071 -0.075 -0.533 -0.465 -0.665 0.173 -0.553 JAWY -0.669 -0.700 -0.401 -0.469 -0.715 -0.038 -0.647 -0.639 LLIPY 0.058 -0.475 0.313 -0.165 -0.388 0.143 0.109 -0.551 COG 0.201 -0.183 0.143 -0.181 -0.029 -0.349 0.024 -0.632 POST -0.149 0.073 -0.088 -0.489 -0.642 -0.570 0.002 -0.481 DurC 0.176 0.738 -0.152 0.664 -0.586 0.309 -0.134 -0.010

Table 4: Regression models computed by the SPSS procedure linear stepwise regression.

Degrees of freedom are always 1,13. The dependent variable is percentage of devoicing and

the independent variables are articulatory positions and EPG measures at the onset of the stop

as well as stop duration.

CONS VP Model R2 F prob Bilabial CG JAWY 0.448 9.73 0.0089

DF DurC 0.545 14.37 0.0026 DurC, JAWY 0.774 18.83 0.0003 RW DurC 0.442 9.51 0.0095 DurC, JAWY 0.611 8.64 0.0056

Velar CG JAWY 0.511 12.53 0.0041 JAWY, JAWX 0.767 18.12 0.0003 DF TDORSY 0.442 9.50 0.0095 JD TDORSX 0.441 9.46 0.0096 TDORSX, JAWX 0.636 9.60 0.0039 RW JAWY 0.408 8.27 0.0139